This isn't going to get easier to deal with. If anything, getting a little absorbed as a vaccine might be part of healthy cognitive hygiene.
The more annoying issue to me is that the kind of sycophantic writing style seems to end up spreading to all LLMs. Maybe it's because they're all optimizing for lmarena or maybe because all those outputs become part of the training data. But the random bolding/italics, creepy-friendly car salesman vibe seems to be getting more common, and custom prompts seem to be getting less effective at keeping them away.